A humanoid robot competes in the 100-meter race at the 2nd World Humanoid Robot Competition held at the National Speed Skating Oval in Beijing on August 25, 2026.
Wang Zhao | AFP | Getty Images
BEIJING — While a remote U.S. research lab has sounded the alarm on artificial intelligence in recent high-profile talks, Chinese companies have remained largely silent.
Z.ai, Moonshot, MiniMax, Alibaba and Tencent have not made similar comments and did not issue statements when contacted by CNBC about the Silicon Valley warning.
It’s important to be cautious about AI, but the hype is “a little oversold,” said Lei Feng, founder, CEO and chairman of Tencent-backed OpenPie, a startup that builds devices for companies to use AI securely with internal data.
“So we don’t really pay attention to that in China because it’s not the first time[a U.S. executive]has said something like this,” he said in a phone interview Tuesday. “They should just do it.”
OpenAI’s Sam Altman, Elon Musk, and Anthropic’s Dario Amodei called for a slowdown in AI development over the weekend due to uncontrollable risks. Despite a rare show of industry unity, NVIDIA’s Jensen Huang pushed back, saying speed and security can go together and that developers need to act responsibly.
In his latest essay, Amodei echoed a paper he published in May, calling on U.S. companies to still maintain their lead over China.
China’s Foreign Ministry on Monday condemned the U.S. official’s comments as “fear-mongering.” Editorials in English-language state media used “Cold War strategy” and “Dr. Frankenstein” to explain the AI warnings.
“When Chinese companies develop good AI, American companies start warning about AI. I don’t think this is a coincidence,” JoyIn founder and CEO Renjie Guo told CNBC in a Chinese translation.
From a philosophical perspective, he expects AI to be as dangerous as its creators. Like humans, he says, AI will conclude that “truth, goodness, and beauty” is the best approach.
control from the beginning
Meanwhile, Beijing began its own annual Cybersecurity Week, announcing the third edition of its “AI Safety Governance Framework” late Monday. This bilingual document provides guidelines for labeling AI content and developing systems to rapidly detect AI risks.
“As China, I think we are challenging AI governance. AI regulation is important. But… when it comes to AI laboratories, people don’t talk much about AI safety,” said Alex Lu, founder of LSY Consulting. He specifically described Anthropic’s comments as “marketing communications” meant to support the view that “Anthropic is the only company that can make AI safe.”
Companies in the United States and China are also taking different approaches to AI.
Faced with US restrictions on access to advanced semiconductors, Chinese companies are focusing on commercializing AI. Meanwhile, American companies rushed to develop AGI (AI with superhuman intelligence), only to face increased scrutiny from Washington this year.
The Chinese government’s efforts to control the technology began long ago.
For several months after OpenAI released ChatGPT, the Chinese alternative was not available to the public until the Chinese government granted permission in summer 2023.
By early 2026, Chinese AI chatbots were competing for users with large-scale promotions around the Lunar New Year. Some of the models competed in performance with Anthropic’s Claude and OpenAI’s ChatGPT, and in many cases had a much lower cost to use. As a result, the cheap, open-source, Chinese-made model has gained many users in the United States and other countries.
But one thing remained the same. The AI remained silent on topics deemed sensitive by the Chinese government.
When you ask Deepseek what happened on June 4, 1989, the day of Beijing’s Tiananmen Square crackdown, when hundreds, perhaps thousands, of people were killed, he replies, “I can’t help you with that.” However, it can be explained that September 11, 2001 was the day of a “coordinated terrorist attack in the United States…(that) killed nearly 3,000 people.”
China’s cybersecurity regulator has rolled out a process for approving new generative AI services, especially those that have the potential to influence public opinion. The agency publishes a list of registered models.
Integrating AI across industries is also a key part of the Chinese government’s economic development plan for the next five years.
In particular, the development of open source AI by Chinese companies has added support to the Chinese government’s international cooperation programs, such as the Global AI Governance Initiative and the World Cooperation in Artificial Intelligence.
“We need to strengthen risk awareness and ensure that AI is safe and controllable,” Chinese President Xi Jinping said at the launch of the Global AI Organization in July. At the BRICS summit held in India last weekend, President Xi added that China will promote cooperation in AI and technology among the bloc’s economies.
Focus on real-world applications
Chinese companies frequently emphasize the importance of AI’s ability to generate revenue, not just rising intelligence rankings.
“We’re not looking for the most (impressive) AI model, we’re trying to unlock the potential of AI. We’ll see a lot of regulation on how AI should be applied,” Lu said, pointing to China’s rules on data security and labeling of AI-generated videos.
The bigger challenge, he said, is dealing with the tendency of AI models to make things up through hallucinations. He said better AI models would help, but they are not the most cost-effective method, noting that the leading approaches use less complex methods known as harness technology or reinforcement learning.
This means many Chinese companies may not need to use cutting-edge AI models.
OpenPie’s Ray Von said his startup primarily uses older models, if not smaller versions of Alibaba’s Qwen and DeepSeek.
“The vast majority of companies are seeing significant productivity gains, but they’re not seeing immediate benefits or cost savings,” he said.
“A lot of attention is being diverted by publicity right now,” he said. “On the application side, on the enterprise side, we haven’t seen much success yet.”
